菜单
Codifies the visual 'ASCII Gearbox' format for debugging agent logic. Used to trace state transitions, logic loops, and evidence grounding.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | logical-gearbox |
| description | Codifies the visual 'ASCII Gearbox' format for debugging agent logic. Used to trace state transitions, logic loops, and evidence grounding. |
Use this format to visualize the 'Internal Gearbox' of a complex discovery/resolution loop.
[PHASE 🔵: EPISTEMIC AUDIT] 🛡️
[PHASE 🟡: THE RATIONALIZER] ⚙️
[PHASE 🟣: INSTRUMENTAL ACTION] 🛠️
[PHASE 🟢: SSoT SYNC] 💾
[PHASE 🔴: THE PARITY SHIELD] 🛡️
Unified c-thru configuration: diagnose the active setup, resolve what a capability alias maps to, switch connectivity modes, remap per-capability models, validate the config, or reload the running proxy. Subcommands: diag [--verbose] | resolve <cap> | mode [<mode>] [--reload] | remap <cap> <model> [--tier <tier>] [--reload] | set-cloud-best-model <cap> <model> [--tier <tier>] [--reload] | set-local-best-model <cap> <model> [--tier <tier>] [--reload] | route <model> <backend> [--reload] | backend <name> <url> [--kind <kind>] [--auth-env <VAR>] [--reload] | agent list | agent set <agent> <cap> [--reload] | agent pin <agent> <model> [--reload] | agent reset <agent> [--reload] | alias list | alias set <pattern> <cap> [--reload] | alias remove <pattern> [--reload] | override list | override set <from> <to> [--reload] | override remove <from> [--reload] | validate | reload | restart [--force]
Agentic infrastructure control for c-thru. Interrogate the proxy for hardware status, active models, and connectivity health. Dynamically switch routing modes (connected, offline, etc.) using natural language prompts.
Iterative review-fix loop: routes each file to the appropriate reviewer (code-reviewer, gas-code-review, gas-ui-review, gas-gmail-cards), applies Fix blocks, rechecks until clean (max 5 rounds), then commits/PRs. commit_mode="pr" (default): commit + push + PR + squash merge + delete branch. commit_mode="commit": commit only (POST_IMPLEMENT pipeline). Git fallback: WIP → last commit when target_files empty. **AUTOMATICALLY INVOKE** after implementing features, fixing bugs, before committing, or after plan implementation completes (user approves + all changes made). **STRONGLY RECOMMENDED** before merging to main, after refactoring, and when code-reviewer returns NEEDS_REVISION. Trigger phrases: "review and fix", "polish this", "clean this up", "make sure this is good", "before committing", "before merging", "loop until clean".
Universal plan review: 3 layers (general quality, code change quality, ecosystem specialization). Invokes gas-plan for GAS plans or node-plan for Node.js/TypeScript plans, conditionally based on detected patterns. AUTOMATICALLY INVOKE when: - MANDATORY_PRE_EXIT_PLAN directive applies (before ExitPlanMode) - User says "review plan", "check plan", "plan ready?" - Any plan file needs review (GAS or non-GAS) NOT for: Code review of existing files (use /gas-review or /review-fix)
Research and update docs/local-model-prompt-techniques.md with latest community findings, Ollama version-specific fixes, and authoritative corrections for all local models in config/model-map.json. Scope: client-observable behavior through the Ollama API — prompting patterns, request parameters, response format, tool-calling behavior, dangers, and best practices. Internal server mechanics and deployment configuration are out of scope unless they directly explain a client-visible API behavior. Spawns parallel research agents per model family, applies a strict citation and confidence standard, and writes a diff-summary of changes. Invoked as /update-model-research [--model <family>] [--section <topic>]
Agentic plan/wave orchestrator. Breaks any task into a wave-based execution plan, drives waves to completion, and tracks state in ${TMPDIR:-/tmp}/c-thru/<repo>/<slug>/. Invoked as /c-thru-plan <user intent>.
